This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
ETA: Energy-based Test-time Adaptation for Depth Completion
Created by
Haebom
Author
Younjoon Chung, Hyoungseob Park, Patrick Rim, Xiaoran Zhang, Jihe He, Ziyao Zeng, Safa Cicek, Byung-Woo Hong, James S. Duncan, Alex Wong
Outline
This paper proposes Energy-based Test-time Adaptation (ETA), a novel method for test-time adaptation of pre-trained depth-completion models. Existing depth-completion models suffer from errors when applied to new data due to covariate shifts caused by environmental changes. ETA explores the data space using adversarial perturbation and trains an energy model without making assumptions about the target data distribution. This energy model evaluates local regions of depth predictions as being within or outside the distribution, and updates the parameters of the pre-trained model to minimize energy at test time, thereby aligning test-time predictions with the source distribution. Experimental results demonstrate an average of 6.94% (outdoor) and 10.23% (indoor) performance improvements over existing state-of-the-art methods on indoor and outdoor datasets.
Takeaways, Limitations
•
Takeaways:
◦
An effective method to improve the test-time adaptive performance of pre-trained depth-completion models is presented.
◦
Building robust models in real-world environments by leveraging adversarial perturbations without assumptions about the target data distribution.
◦
Significant performance improvements over previous best-in-class performance on indoor and outdoor datasets.
•
Limitations:
◦
The training and testing time adaptation process of energy models can be computationally expensive.
◦
Additional evaluation of generalization performance across various environmental changes is needed.
◦
Further research is needed to further improve the effectiveness of the proposed method.
